Zha0q1 opened a new pull request #18925: URL: https://github.com/apache/incubator-mxnet/pull/18925
This PR fixes numpy dot on large tensors. This fix also requires ILP 64 Blas which CI has not migrated to yet, so for now we are skipping the test. With this fix numpy dot will no longer error out on > 2**32 dimensions. However there seems to be a precision issue with ILP64 openblas: https://github.com/xianyi/OpenBLAS/issues/2779 So on my remote computer ``` A = np.ones((1, 2**31)) B = np.ones((2**31, 1)) C = np.dot(A, B) ``` Will give 1.93274e+09 ``` A = np.ones((2**31, 1), dtype='float64') B = np.ones((1, 2**31), dtype='float64') ``` Will give the arithmetically correct result 2.14748365e+09 ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
